Modeling and Feedback Design for Autonomic Management of Enterprise-scale Virtualized Data Centers

This project aims to develop control-theoretic modeling techniques, feedback design methodologies, and virtual machine management tools for resource provisioning in enterprise-scale data centers. First, the Linear-Parameter-Varying approach will be applied to characterize the time-varying behavior of traffic and to provide system adaption to dynamical variation of workloads and operating conditions. Then a distributed yet coordinated control architecture will be developed to address time-varying resource bottlenecks across different tiers for multi-tier applications and to cope with the large scale and complexity of modern data centers. In the end, this project will develop virtual machine implementation of the proposed resource management techniques and prototype a virtualized data center to evaluate the performance of the proposed modeling and design algorithms. If successful, the results of this research will make the following contributions to the computer systems scientific-community and industries: 1) performance models of server systems that can cope with dynamically changing workloads with complex nature, and more importantly, a unifying paradigm that allows the combination of control-theoretic approaches with new advances in workload characterization in the computer systems community; and 2) control architecture that can deal with the growing scale and complexity of modern data centers. In addition, the results of this research will be transformable to ready-to-use software and products for data centers through our collaboration with industrial sponsors. In education, the integrated research and education program will teach fundamentals of control theory in nontraditional control domains, outreach for high school students, and promote diversity by engaging women and minorities.